4 ms·
A more useful system would take Opus-compressed data as input and feature-extract that, presumably faster than this thing. Bonus for not requiring a proprietary
by ncmncm 6y ago
A more useful system would take Opus-compressed data as input and feature-extract that, presumably faster than this thing. Bonus for not requiring a proprietary library like libsparse_inference.so.
Also, instead of encoding independent 40ms segments, it should be much better to encode 10ms segments given the previous 30ms.